fnhum-11-00399 July 28, 2017 Time: 16:14 # 1

ORIGINAL RESEARCHpublished: 02 August 2017

doi: 10.3389/fnhum.2017.00399

Edited by:Carol Seger,

Colorado State University,United States

Reviewed by:Anthony Joseph Ryals,

University of North Texas,United States

Anne-Pascale Le Berre,Stanford University Schoolof Medicine, United States

*Correspondence:Liang Luo

luoliang@bnu.edu.cnWenjing Wang

wangwenjing@bnu.edu.cnChongde Lin

linchongde@263.net

Received: 14 May 2017Accepted: 19 July 2017

Published: 02 August 2017

Citation:Hu X, Liu Z, Chen W, Zheng J,

Su N, Wang W, Lin C and Luo L(2017) Individual Differences

in the Accuracy of Judgmentsof Learning Are Related to the Gray

Matter Volume and FunctionalConnectivity of the Left Mid-Insula.

Front. Hum. Neurosci. 11:399.doi: 10.3389/fnhum.2017.00399

Individual Differences in theAccuracy of Judgments of LearningAre Related to the Gray MatterVolume and Functional Connectivityof the Left Mid-InsulaXiao Hu1,2, Zhaomin Liu3, Wen Chen2, Jun Zheng1,2, Ningxin Su1,2, Wenjing Wang2*,Chongde Lin4* and Liang Luo1,2,4*

1 Collaborative Innovation Center of Assessment Toward Basic Education Quality, Beijing Normal University, Beijing, China,2 State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China, 3 School ofSociology, China University of Political Science and Law, Beijing, China, 4 Institute of Developmental Psychology, BeijingNormal University, Beijing, China

The judgment of learning (JOL) is an important form of prospective metamemoryjudgment, and the biological basis of the JOL process is an important topic inmetamemory research. Although previous task-related functional magnetic resonanceimaging (MRI) studies have examined the brain regions underlying the JOL process, theneural correlates of individual differences in JOL accuracy require further investigation.This study used structural and resting-state functional MRI to investigate whetherindividual differences in JOL accuracy are related to the gray matter (GM) volume andfunctional connectivity of the bilateral insula and medial Brodmann area (BA) 11, whichare assumed to be related to JOL accuracy. We found that individual differences in JOLaccuracy were related to the GM volume of the left mid-insula and to the functionalconnectivity between the left mid-insula and various other regions, including the leftsuperior parietal lobule/precuneus, bilateral inferior parietal lobule/intraparietal sulcus,right frontal pole and left parahippocampal gyrus/fusiform gyrus/cerebellum. Furtheranalyses indicated that the functional connectivity related to individual differences in JOLaccuracy could be divided into two factors and might support information integrationand selective attention processes underlying accurate JOLs. In addition, individualdifferences in JOL accuracy were not related to the GM volume or functional connectivityof the medial BA 11. Our findings provide novel evidence for the role of the left mid-insulaand its functional connectivity in the JOL process.

Keywords: functional connectivity, gray matter volume, insula, judgments of learning, metamemory

INTRODUCTION

Metamemory refers to the processes of monitoring and controlling memory activities (Nelsonand Narens, 1990). One important form of metamemory is prospective metamemory judgment,which refers to prediction of future memory performance (Dunlosky and Metcalfe, 2009). Theaccuracy of prospective metamemory judgments is typically evaluated by examining the extent to

Frontiers in Human Neuroscience | www.frontiersin.org 1 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 2

Hu et al. Judgments of Learning and Insula

which prospective metamemory judgments discriminate betweenremembered and forgotten items (Dunlosky and Metcalfe, 2009).If the subjective level of prospective metamemory judgmentsis higher for remembered items than forgotten items, then theaccuracy of prospective metamemory judgments is high. Incontrast, prospective metamemory accuracy is low if people givehigher judgments to forgotten items. Prospective metamemoryjudgments can be made either at the stage of acquiring knowledge(called judgment of learning, or JOL) or at the time of retrieval(called feeling of knowing, or FOK) (Nelson and Narens, 1990).JOL is an important type of prospective metamemory judgment,and previous studies have indicated that JOL accuracy is essentialfor appropriate guidance of subsequent learning and memoryprocesses (Dunlosky and Metcalfe, 2009; Bjork et al., 2013).People with more accurate JOLs can use strategies that are moreappropriate in subsequent learning processes, such as allocatingstudy time more appropriately during self-regulated learning orchoosing more important and valuable items for restudy (Nelsonand Narens, 1990; Dunlosky and Metcalfe, 2009). Given theimportance of JOLs in learning processes, the biological basis ofthe JOL process is an important topic in metamemory research(Schwartz and Bacon, 2008; Chua et al., 2014).

Some lesion-based studies have investigated whether frontallobe lesions significantly affect JOL accuracy (Vilkki et al., 1998,1999). For example, Vilkki et al. (1998) asked healthy participantsand patients with brain lesions to learn a list of words. Beforethe recall test, all participants had to make a JOL to predicthow many words they could recall in the test. They found thatpatients with left or right frontal lobe lesions had significantlylower JOL accuracy than the control group. In Vilkki et al. (1999),patients and healthy control participants were required to learnthe locations of different faces and make a JOL before the test.The results showed that patients with damaged right frontal lobesshowed decreased JOL accuracy. These studies suggest that thefrontal lobe may play an important role in the JOL process.However, Vilkki et al. (1998, 1999) did not identify which partof the frontal lobe is essential to JOL accuracy.

In addition to lesion-based studies, functional magneticresonance imaging (MRI) studies have investigated the brainregions underlying the JOL process (Kao et al., 2005; Do Lamet al., 2012; Yang et al., 2015). An early study by Kao et al.(2005) found that the left lateral frontal cortex, left ventromedialprefrontal cortex and several temporal, parietal, occipital andlimbic regions were associated with predicted encoding success(i.e., higher activation for high-JOL trials than for low-JOLtrials). They also examined the relationship between individualdifferences in JOL accuracy and the activation of the regionsrelated to predicted encoding success, and they found thatgreater JOL accuracy was correlated with higher activation inthe ventromedial prefrontal cortex. Another study by Do Lamet al. (2012) separated JOL trials from initial encoding andindicated that the medial prefrontal cortex, medial orbitofrontalcortex and anterior cingulate cortex were more active duringhigh-JOL trials. In addition, Yang et al. (2015) compared theneural activities for encoding trials subsequently given highand low JOLs, and found that the dorsolateral, rostrolateraland ventromedial prefrontal cortex, posterior cingulate cortex,

middle temporal gyrus, superior lateral occipital cortex andangular gyrus showed subsequent JOL effects (i.e., higheractivation for trials subsequently given high JOLs than thosegiven low JOLs). They also found that greater JOL accuracywas correlated with smaller subsequent JOL effects in theventromedial prefrontal cortex. In these previous studies, theonly region consistently involved in the JOL process was theventromedial prefrontal cortex, particularly the medial part ofBrodmann area (BA) 11. Similarly, a study of FOK has indicatedthat the activation of medial BA 11 is also related to accurateFOK judgments (Schnyer et al., 2005), suggesting that medialBA 11 is essential to different types of prospective metamemoryjudgments.

Another important method for investigating the neuralcorrelates of metamemory processes is to examine therelationship between individual differences in brain structureand metamemory accuracy. To our knowledge, no study hasexamined the relationship between JOL accuracy and brainstructure. However, researchers have used structural MRI toinvestigate the relationship between the FOK process andthe gray matter (GM) volume in the brain (Cosentino et al.,2015; Le Berre et al., 2016). For example, Le Berre et al. (2016)asked alcoholic patients to make FOK judgments in the recallphase about whether they could recognize the target word ina subsequent recognition test; they found that greater FOKaccuracy was correlated with higher mean GM volume of thebilateral insula. Similarly, Cosentino et al. (2015) indicated thathigher mean GM volume of the right insula was related to higherFOK accuracy in older adults. Both studies have shown thatindividual differences in FOK accuracy are related to the GMvolume of the insula.

Although previous studies have investigated the neuralcorrelates of the JOL process from different perspectives, twoimportant issues require further examination. First, although theactivation of medial BA 11 has been shown to be consistentlyassociated with the JOL process (Kao et al., 2005; Do Lamet al., 2012; Yang et al., 2015), no study has examined whetherindividual differences in JOL accuracy are related to the GMvolume of medial BA 11. Second, previous structural MRIstudies have not investigated whether JOL is also related tothe GM volume of the insula, as found in structural MRIstudies concerning FOK (Cosentino et al., 2015; Le Berre et al.,2016). Researchers have suggested that in FOK processes, theinsula may be related to the monitoring of task performance(Cosentino et al., 2015) or self-inferential processes to generateaccurate future estimations (Le Berre et al., 2016). Thesecognitive processes are also involved in the JOL process (Koriat,1997; Dunlosky and Metcalfe, 2009). Thus, it is possible thatthe GM volume of the insula may also be related to JOLaccuracy. Investigating these issues could further reveal theneural correlates of the JOL process. To address these issues, inthe present study, we first investigated the relationship betweenJOL accuracy and GM volume in the brain.

In addition, previous neuroimaging studies have mainlyfocused on the particular brain regions related to JOL processand have not examined whether JOLs are correlated with thefunctional connectivity between different regions. According to

Frontiers in Human Neuroscience | www.frontiersin.org 2 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 3

Hu et al. Judgments of Learning and Insula

the widely accepted cue-utilization theory, JOLs are inferentialin nature and are related to the processing and integrationof various cues (e.g., study time, processing fluency) relevantto study materials. When making JOLs, people first monitorand process the cues that may contribute to the memoryprocess and then integrate these cues to generate estimationsfor future memory performance (Koriat, 1997; Dunlosky andMetcalfe, 2009). Thus, it is possible that JOLs are associatedwith the coupling between different regions that are related tothe processing and integration of different cues. In addition, aprevious study suggests that retrospective confidence judgment(RCJ), another type of metamemory judgment made after amemory test, is related to functional connectivity in the brain(Baird et al., 2013). For example, Baird et al. (2013) indicated thatindividual differences in the accuracy of retrospective confidencein a memory test were significantly correlated with the functionalconnectivity between the anterior prefrontal cortex, in whichthe brain structure is related to the accuracy of RCJs (Fleminget al., 2010), and several parietal regions during the resting state.Based on the results concerning RCJs, we speculated that JOLsmight also be related to functional connectivity between differentregions. Furthermore, making JOLs is similar to the decision-making process (Schwarz, 2004), which is also related to theintegration of different information (Bettman et al., 1998; Limet al., 2013). A previous study has suggested that the evaluationprocess in decision making is associated with the functionalconnectivity between several regions of the temporal cortex(which are related to the processing of different information)and the prefrontal cortex (which is related to the integration ofinformation) (Lim et al., 2013). Thus, JOLs may also be associatedwith the functional connectivity between various brain regions.The second purpose of the present study was to investigate therelationship between JOL accuracy and functional connectivityin the brain.

Taken together, the main goals of the present study were(1) to examine the relationship between individual differencesin JOL accuracy and the GM volume in the brain and (2) toinvestigate the functional connectivity associated with individualdifferences in JOL accuracy. To achieve the first goal, we analyzedthe correlation between individual differences in JOL accuracyand the GM volume. In particular, we examined the correlationbetween JOL accuracy and the GM volume of medial BA 11and the insula. Medial BA 11 is consistently involved in the JOLprocess (Kao et al., 2005; Do Lam et al., 2012; Yang et al., 2015)and has been correlated with JOL accuracy in previous task-related fMRI studies (although the direction of the correlationis controversial; see Kao et al., 2005; Yang et al., 2015). Thus,the GM volume of medial BA 11 may also be correlated withJOL accuracy. In addition, FOK studies have consistently revealedpositive correlations between FOK accuracy and the GM volumeof the insula (Cosentino et al., 2015; Le Berre et al., 2016).According to the discussion above, some cognitive processes thatunderlie FOK and may be related to the insula are also involvedin the JOL process (Koriat, 1997; Dunlosky and Metcalfe, 2009),and it is possible that the GM volume of the insula might alsobe related to JOL accuracy. We hypothesized that JOL accuracymight show positive correlation with the GM volume of the

insula (similar to FOK studies) and might also show significantcorrelation with the GM volume of medial BA 11 (althoughthe direction of the correlation is unknown). There are twocommonly used methods to analyze the relationship between GMvolume and individuals’ performance: the traditional region-of-interest (ROI) approach used in previous FOK studies (Cosentinoet al., 2015; Le Berre et al., 2016) and voxel-based morphometry(VBM) analysis. In the ROI-based approach, the correlationsbetween individuals’ performance and the mean GM volumewithin the ROIs are analyzed. By contrast, VBM analysispermits the calculation of the correlation between individuals’performance and GM volume voxel by voxel (Luders et al., 2004;Bhaskar et al., 2013). One pitfall of the ROI-based approach isthat it is used primarily on large areas, and differences in GMvolume in small parts of the ROIs may therefore be overlooked.By contrast, VBM analysis can examine focal differences inbrain anatomy and reveal specific clusters related to individuals’performance (Bhaskar et al., 2013). Thus, in this study, wechose VBM analysis to evaluate the relationship between GMvolume and JOL accuracy. We conducted VBM analysis in amask containing the bilateral insula and medial BA 11 to examinewhether JOL accuracy was significantly correlated with the GMvolume in these regions.

To achieve the second goal, we used resting-state fMRIto examine whether individual differences in JOL accuracyare related to the resting-state functional connectivity betweendifferent brain regions. Resting-state functional connectivity isindexed by correlations in low-frequency fluctuations of theresting-state fMRI signal (Lee et al., 2013). In contrast totask-related brain activity, which is evoked by specific taskstimuli, resting-state functional connectivity may characterizethe intrinsic functional organization of the brain (Fox andRaichle, 2007) and has high reliability and reproducibility (Leeet al., 2013). Although resting-state functional connectivity is notdirectly related to any cognitive task, a previous study suggeststhat individual differences in resting-state functional connectivitycan predict task-induced brain activity (Mennes et al., 2010). Inaddition, resting-state functional connectivity has been shown tobe closely related to individual differences in various cognitivefunctions, such as language, decision making and intelligence(Song et al., 2008; Li et al., 2013; Xiao et al., 2016). In the resting-state analysis, we used the significant cluster found in VBManalysis as a seed region and conducted whole-brain seed-basedfunctional connectivity analysis. Using the cluster found in VBManalysis as the seed region is a widely accepted method in resting-state analysis (Loh and Kanai, 2014; Evans et al., 2015). Wehypothesized that if the VBM analysis revealed that the clustersin medial BA 11 or insula were significantly correlated with JOLaccuracy, the functional connectivity between these clusters andother regions might also support the process of making accurateJOLs. In addition, to further explore the relationship betweenJOL accuracy and functional connectivity in the brain, weconducted an exploratory ROI-wise analysis to examine whetherJOL accuracy was correlated with the functional connectivitybetween regions connected to the seed region in the seed-based analysis. We also performed exploratory factor analysison the functional connectivity found in the analyses above.

Frontiers in Human Neuroscience | www.frontiersin.org 3 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 4

Hu et al. Judgments of Learning and Insula

Factor analysis is a statistical technique that can determine whichvariables form coherent subsets (factors) that are somewhatindependent. Factors are hypothesized to reflect latent constructsor underlying processes that result in correlations betweenvariables (Fabrigar et al., 1999). Thus, factor analysis may providea better understanding of the functions of the resting-statenetwork related to JOL accuracy by revealing the relationshipsbetween functional connectivity outcomes (Jacobs et al., 2014).

MATERIALS AND METHODS

ParticipantsThirty-five students (13 males, 22 females; aged 19–28 years)participated in this study. The participants were right-handed,had normal or corrected-to-normal vision, and reported nopersonal or family history of neurological or psychiatricdisorders. This study was carried out in accordance with therecommendations of the Institutional Review Board of theState Key Laboratory of Cognitive Neuroscience and Learning,Beijing Normal University with written informed consent fromall subjects. All subjects gave written informed consent inaccordance with the Declaration of Helsinki.

MaterialsThe materials consisted of 300 two-character Chinese wordsfrom the Chinese word database of Cai and Brysbaert (2010).The word frequency varied between 0.24 and 46.29 per millionwords. The 300 words were randomly divided into two sets (eachcontaining 150 words). The two sets of words did not differin word frequency, concreteness, familiarity or the number ofstrokes (p > 0.05). One set of words was presented during theencoding phase, and the other set was used as new words duringthe recognition phase. The assignment of the two sets to theencoding and recognition phases was counterbalanced acrossparticipants.

Task and ProcedureThe participants completed two experimental sessions: abehavioral session that included a memory task and an MRIsession in which T1-weighted structural images and resting-statefMRI images were acquired. The MRI data were acquired beforethe behavioral experiment.

The memory task consisted of two phases, encoding andrecognition, with 24 h between the two. On the first day, theparticipants completed the encoding phase. Before beginning theencoding phase, the participants were informed that there wouldbe a recognition test after 24 h in which their memory of thepresented words would be tested by asking that they select thepreviously presented word from two choices (as described below).During the encoding phase, the participants sequentially viewed150 words that were presented in the center of a screen, witheach word presented for 2 s. The words were presented in arandom order. Immediately following the presentation of eachword, the participants were instructed to rate the probability thatthey would correctly choose that word during the recognitiontest. Ratings were made using a 6-point Likert-type scale ranging

from 1 (very unlikely) to 6 (very likely) and were entered with thenumber pad of a keyboard.

The participants left the lab after finishing the encoding phase.They then returned 24 h later to participate in the recognitiontest, which used a two-alternative forced choice (2AFC) designand contained 150 trials. In each recognition trial, two wordswere presented on the left and right of the screen. One of thesewords had been presented in the encoding phase (“old”), whilethe other word had not been presented previously (“new”). Theparticipants had to press a key to indicate which word was old andthen rate their confidence in the accuracy of their response on a 6-point scale ranging from 1 (very low confidence) to 6 (very highconfidence). The old and new words were presented in randomorder, and the location (left or right) of the old words was alsorandomized.

The memory task procedure is shown in Figure 1. Stimuluspresentation and response collection were controlled withE-Prime software (Schneider et al., 2002). The confidence ratingsmade during the recognition phase were not analyzed in thisstudy.

Quantification of Memory Performanceand JOL AccuracyWe used the proportion of correctly answered trials inthe recognition test to quantify the participants’ memoryperformance. JOL accuracy was quantified by calculating thearea under the receiver operating characteristic curve (AUROC)(Beaman et al., 2014; Ryals et al., 2016). Previous studies havemainly used two measures of JOL accuracy: Goodman–Kruskal’sgamma and the AUROC. As a traditional measure of JOLaccuracy, gamma has been used in behavioral studies and someneuroimaging studies (Nelson, 1984; Kao et al., 2005). However,gamma is susceptible to the tendency to make higher or lowerjudgment ratings (response bias), which can lead to erroneousinterpretations of JOL accuracy (Masson and Rotello, 2009).By contrast, as a non-parametric measure based on Type IIsignal detection theory (SDT), the AUROC is not influenced byresponse bias and is a more accurate measure of JOL accuracythan gamma (Ryals et al., 2016). In addition, the AUROC is alsoused to measure the accuracy of other forms of metamemory,such as RCJs (Ryals et al., 2016; Valk et al., 2016).

The AUROC was determined according to previouslypublished methods (Fleming et al., 2010; Fleming and Lau, 2014)by calculating the area under the Type II receiver operatingcharacteristic (ROC) curve using non-parametric methods. Weconstructed the ROC curve by treating each JOL level as acriterion that separated high JOLs from low JOLs. For example,we started with a liberal criterion that assigned low JOL = 1and high JOL = 2–6, then a higher criterion that assignedlow JOL = 1–2 and high JOL = 3–6, and so on. For eachsplit of the data, the hit rate hi = p (high JOL | correct) andfalse alarm rate f i = p (high JOL | incorrect) were calculatedand used to construct an x-y point on the ROC curve. TheROC curves were anchored at [0,0] and [1,1]. According toFleming et al. (2010), an ROC curve that bows sharply upwardindicates that the probability of being correct increases rapidlywith confidence. Conversely, a flat ROC function indicates a

Frontiers in Human Neuroscience | www.frontiersin.org 4 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 5

Hu et al. Judgments of Learning and Insula

FIGURE 1 | The experimental paradigm of the memory task. The old/new labels were not presented in the recognition phase. They are presented here to explain theexperimental procedure.

weak link between confidence and accuracy. Application ofthe Kolmogorov–Smirnov test revealed that the AUROC wasnormally distributed (p > 0.2).

To explore how well the Type II SDT model accounted forthe JOL rating data, we fit the following linear regression model(Fleming et al., 2010):

z(h) = β0 + β1z(f) + ε

where z is the inverse of the cumulative normal distributionfunction. This model provided an excellent fit to the data (meanR2= 0.97).Some researchers have noted that the AUROC may be

confounded by task performance (Galvin et al., 2003). Thus, inthe subsequent structural MRI and resting-state fMRI analyses,recognition accuracy was included as a covariate (Valk et al.,2016).

Structural MRIData AcquisitionThe structural MRI data were acquired using a 3T Siemens TimTrio MRI scanner at the Imaging Center for Brain Research,Beijing Normal University. Using a magnetization preparedrapid gradient echo (MPRAGE) sequence, high-resolutionT1-weighted structural images (TR = 2530 ms, TE = 3.39 ms,inversion time = 1100 ms, flip angle = 7◦, 144 sagittalslices, FOV = 192 mm × 256 mm × 256 mm, voxelsize= 1.33 mm× 1 mm× 1 mm) were acquired.

Voxel-Based Morphometry (VBM) AnalysisStructural brain images were analyzed with the VBM8 toolbox1,which was incorporated into SPM8 software2 running onMATLAB R2010b (MathWorks, Natick, MA, United States).Preprocessing was completed using the default settings of VBM8.In brief, the following steps were performed: (1) intrasubjectbias correction; (2) segmentation into different tissue classes; (3)linear and non-linear registration to the Montreal Neurological

1http://dbm.neuro.uni-jena.de/vbm2http://www.fil.ion.ucl.ac.uk/spm/software/spm8/

Institute (MNI) space (resliced to 1.5 mm × 1.5 mm × 1.5 mm);and (4) modulation of tissue segments using non-linearnormalization parameters to account for individual differencesin brain size. The normalized GM segments were then smoothedusing an 8-mm full-width at half maximum (FWHM) Gaussiankernel.

Statistical analysis was performed using SPM8. We used amask that contained the bilateral insula and medial BA 11 asthe mask in the statistical analysis to examine whether the GMvolume in these regions was related to individual differences inJOL accuracy. The bilateral insula region was derived from thexjView toolbox3. The BA 11 region was also created from thexjView toolbox, and we used the medial part (−20 < × < 20)of BA 11 in the following analysis (Hogeveen et al., 2016). Theparticipants’ age, gender, recognition accuracy and total volumeof brain tissue were included as covariates. Voxels with GMvalues of <0.2 (absolute threshold masking) were excluded fromthe analysis to avoid possible edge effects between differenttissue types. We applied a height threshold of p < 0.001 anda cluster-level threshold of p < 0.05 using both the family-wise error (FWE) correction and the non-stationary clustercorrection implemented in SPM8 to account for the non-isotropic smoothness of the VBM data (Hayasaka et al., 2004).In addition, we conducted a whole-brain VBM analysis using thesame threshold to examine whether JOL accuracy was correlatedwith the GM volume of other regions in the brain.

We also examined whether the GM volume of the regionidentified in the analysis described above was significantlycorrelated with the participants’ recognition performance. Themean GM volume of the region identified in the VBM analysiswas extracted, and we calculated the correlation between the GMvolume of the region and the participants’ recognition accuracy.

Resting-State fMRIData AcquisitionThe resting-state fMRI data were acquired using a 3TSiemens Tim Trio MRI scanner at the Imaging Center

3http://www.alivelearn.net/xjview/

Frontiers in Human Neuroscience | www.frontiersin.org 5 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 6

Hu et al. Judgments of Learning and Insula

for Brain Research, Beijing Normal University. Resting-state functional MRI images were obtained using anecho-planar sequence sensitive to blood oxygenation level-dependent contrast (TR = 2000 ms, TE = 30 ms, flipangle = 90◦, 33 axial slices acquired interleaved with a 0.7-mm gap, voxel size = 3.125 mm × 3.125 mm × 4.2 mm,FOV = 200 mm × 200 mm × 138.6mm, 200 volumes). Theparticipants were instructed to stay awake and to keep their eyesclosed during the functional runs.

Data PreprocessingPreprocessing of the resting-state fMRI data was performedusing the DPARSF toolbox4 (Yan and Zang, 2010). The first10 volumes were removed to account for the T1 equilibriumeffect, leaving 190 volumes for the final analysis. The functionalimages were sinc interpolated in time to correct for the slicetime differences. Then, the images were realigned with a rigidbody linear transformation to correct for head movements. Next,the functional images were co-registered with the correspondingT1 volume and warped into MNI space at a resolution of3 mm × 3 mm × 3 mm using Diffeomorphic AnatomicalRegistration Through Exponentiated Lie (DARTEL) algebrain SPM8 (Ashburner, 2007). The images were then spatiallysmoothed using a 6-mm FWHM Gaussian filter to reduce noise,and the nuisance covariates were regressed out. The nuisancecovariates included 24 motion parameters that were calculatedfrom the six original motion parameters using Volterra expansion(Friston et al., 1996). These parameters have been shown tobe better at decreasing motion effects than the six originalparameters alone (Yan et al., 2013). The white matter andcerebrospinal fluid signals were also covaried out, and lineardetrending was performed. The global signal was not included asa nuisance covariate because recent work suggests that regressionof the global signal may reduce the accuracy of the connectivityestimates (Saad et al., 2012). The images were then filtered at0.01–0.1 Hz.

Seed-Based Functional Connectivity AnalysisIn the VBM analysis, we found that individual differences inJOL accuracy were related to GM volume in a cluster of the leftmid-insula (see the Results section). Thus, the first seed regionfor the seed-based functional connectivity analysis was the leftmid-insula identified in the VBM analysis. Although we didnot find a relationship between GM volume in medial BA 11and JOL accuracy (see the Results section), we used the medialBA 11 region defined in the VBM analysis as the second seedregion to investigate whether JOL accuracy was correlated withthe functional connectivity of medial BA 11. The mean timecourse of all voxels in each seed region was used to calculatethe voxel-wise linear correlations (Pearson’s r) throughout thewhole brain. The connectivity maps were transformed from r toz values using Fisher’s z transformation and then submitted tosecond-level analyses in SPM8. These analyses examined whetherthe functional connectivity of each seed region was related tothe participants’ JOL accuracy. For the connectivity map of each

4http://www.restfmri.net/forum/DPARSF

seed region, we calculated the correlation between participants’JOL accuracy and the z-transformed correlation value of eachvoxel. The participants’ age, gender and recognition accuracywere included as covariates. Similar to previous study (Bairdet al., 2013), in the whole-brain analysis, we applied a heightthreshold of p < 0.005 and a cluster-level threshold of p < 0.05with false discovery rate (FDR) correction. In addition to thewhole-brain analysis, we also examined whether JOL accuracywas related to the functional connectivity between the left mid-insula and medial BA 11. We conducted a seed-based functionalconnectivity analysis in the medial BA 11 mask when using theleft mid-insula as the seed region and in the left mid-insula maskwhen using medial BA 11 as the seed region.

We also examined whether the functional connectivity foundin the analysis described above was significantly correlatedwith the participants’ recognition performance. The meanz-transformed correlation value for each functional connectivityidentified in the analysis above was extracted, and we calculatedthe correlation between the mean z-transformed correlationvalues of each functional connectivity and the participants’recognition accuracy. In addition, we investigated whether therelationship between JOL accuracy and functional connectivityfound in the analysis above was solely due to the difference in theGM volume of the seed region.

ROI-Wise Functional Connectivity AnalysisIn the whole-brain seed-based functional connectivity analysis,we found that JOL accuracy was significantly related to thefunctional connectivity between the left mid-insula and five otherregions (see the Results section). To further explore whetherindividual differences in JOL accuracy were correlated with thefunctional connectivity between the regions connected to the leftmid-insula, we conducted an exploratory ROI-wise analysis usingthe five regions discovered in the seed-based analysis as ROIs. Foreach participant, we extracted the mean time series by averagingacross all voxels in each ROI and then computed bivariatecorrelation coefficients for each pair of ROIs. The resultantROI-to-ROI correlation values were Fisher z-transformed. Wecalculated the partial correlation between JOL accuracy and thez-transformed correlation of each ROI-to-ROI pair, controllingfor age, gender and recognition accuracy. To correct for multiplecomparisons, we applied a threshold of FDR-corrected p < 0.05for the partial correlation analyses (Benjamini and Hochberg,1995). We also examined whether the ROI-to-ROI functionalconnectivity that was related to JOL accuracy was significantlycorrelated with the participants’ recognition performance.

Exploratory Factor AnalysisThe seed-based analysis and ROI-wise analysis revealed a brainnetwork with functional connectivity of nine ROI-to-ROI pairsrelated to JOL accuracy (see the Results section). To examinewhether the functional connectivity found in the analyses abovecould form different subsets, we conducted an exploratory factoranalysis (Jacobs et al., 2014). For each participant, we extractedthe z-transformed correlation of each ROI-to-ROI pair that wassignificantly correlated with JOL accuracy. The number of factorsretained was determined using principal component analysis as

Frontiers in Human Neuroscience | www.frontiersin.org 6 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 7

Hu et al. Judgments of Learning and Insula

an extraction method and a threshold of eigenvalue >1, followedby oblique rotation (Promax). We then conducted a three-steplinear regression analysis to investigate whether each of thefactors could predict the participants’ JOL accuracy.

RESULTS

Behavioral ResultsOn average, the participants’ recognition accuracy (theproportion of correctly answered trials during the recognitiontest) was 0.823 (SD = 0.092, range 0.59–0.96), which wassignificantly higher than the chance level (0.5), t(34) = 20.7,p < 0.001, Cohen’s d = 3.50. We then calculated the AUROCof the JOL ratings for each participant. The mean AUROC forall participants was 0.551 (SD = 0.073, range 0.371–0.716),which was also significantly higher than the chance level (0.5),t(34) = 4.13, p < 0.001, Cohen’s d = 0.70. These resultsindicate that participants could significantly predict theirfuture memory performance, although their JOL accuracy wasrelatively low. In addition, there was no significant correlationbetween recognition accuracy and the AUROC, r (35) =−0.010,p = 0.957, suggesting that JOL accuracy may be independentof memory performance. To investigate whether the AUROCwas actually independent of the response bias in the JOL ratings,we calculated the Type II bias for each participant according tothe method of a previous study (Kornbrot, 2006). The resultsshowed that the AUROC and Type II bias were not significantlycorrelated, r(35) = −0.038, p = 0.829, and thus confirmed thatthe AUROC was not influenced by response bias. In addition,we calculated the correlation between the AUROC and the headmotion measurement (mean framewise displacement) in theresting-state fMRI data (Power et al., 2012) and found that thiscorrelation was not significant, r(35)=−0.120, p= 0.491.

VBM Analysis ResultsWe correlated JOL accuracy with the GM volume of each voxelin the mask containing the bilateral insula and medial BA 11and found that greater JOL accuracy was only correlated withhigher GM volume of a cluster in the left mid-insula (BA 13;peak MNI: −40.5, 0, 15; see Figure 2 and Table 1). The volumeof this left mid-insula cluster (1006 mm3) accounted for 4% ofthe volume in the left insula (24960 mm3) and for 1.3% of thevolume in the whole mask (74944 mm3). JOL accuracy did notshow a negative correlation with the GM volume of the insula ormedial BA 11. In addition, whole-brain analysis did not reveal anyregion in which the GM volume was significantly correlated withJOL accuracy. We also examined whether the GM volume of theleft mid-insula was significantly correlated with the participants’recognition performance and found that this correlation was notsignificant (p > 0.9).

Seed-Based Functional ConnectivityAnalysis ResultsWe then conducted whole-brain functional connectivity analysisto investigate whether the functional connectivity of the left

FIGURE 2 | Gray matter (GM) volume of the left mid-insula was related toindividual differences in judgment of learning (JOL) accuracy.

mid-insula, which was identified in the VBM analysis, andmedial BA 11 was related to the participants’ JOL accuracy.As shown in Table 1 and Figure 3, greater JOL accuracywas significantly correlated with greater functional connectivitybetween the left mid-insula and five clusters: the left superiorparietal lobule/precuneus (SPL/Pcu, BA 7; peak MNI: −12, −66,63), left inferior parietal lobule/intraparietal sulcus (IPL/IPS, BA40; peak MNI: −45, −36, 42), right IPL/IPS (BA 40; peak MNI:45, −39, 51), right frontal pole (BA 10; peak MNI: 39, 51, 21),and a cluster (peak MNI: −27, −21, −30) including part of theleft parahippocampal gyrus (BA 36), left fusiform gyrus (BA 20),and left cerebellum. JOL accuracy did not show a significantnegative correlation with the functional connectivity between theleft mid-insula and any region. In addition, JOL accuracy was notsignificantly correlated with the functional connectivity betweenmedial BA 11 and any region. To examine whether JOL accuracywas related to the functional connectivity between the left mid-insula and medial BA 11, we conducted seed-based functionalconnectivity analysis in the medial BA 11 mask when using theleft mid-insula as the seed region and in the left mid-insula maskwhen using medial BA 11 as the seed region. The results revealedthat JOL accuracy did not show a significant correlation with thefunctional connectivity between the left mid-insula and medialBA 11.

We also examined whether the functional connectivity foundin the analysis described above was significantly correlatedwith the participants’ recognition performance and found thatno functional connectivity was significantly correlated withrecognition accuracy (all p > 0.3).

One possibility is that the correlation between JOL accuracyand functional connectivity might be due to differences in the

Frontiers in Human Neuroscience | www.frontiersin.org 7 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 8

Hu et al. Judgments of Learning and Insula

TABLE 1 | Brain areas positively correlated with judgment of learning (JOL) accuracy in voxel-based morphometry (VBM) and whole-brain seed-based functionalconnectivity analyses.

Anatomical areas BA Volume (mm3) MNI coordinates t-value

X Y Z

VBM analysis

L. mid-insula 13 1006 −40.5 0 15 4.31

Seed-based functional connectivity analysis

L. superior parietal lobule 7 3240 −12 −66 63 5.01

L. Precuneus 7 −18 −48 57 4.77

L. parahippocampal gyrus 36 2241 −27 −21 −30 4.33

L. fusiform gyrus 20 −48 −27 −27 4.11

L. cerebellum −30 −33 −36 4.11

L. inferior parietal lobule/intraparietal sulcus 40 5670 −45 −36 42 4.25

R. frontal pole 10 2619 39 51 21 4.09

R. inferior parietal lobule/intraparietal sulcus 40 2214 45 −39 51 3.75

L, left; R, right; BA, Brodmann’s area.

FIGURE 3 | Whole-brain functional connectivity of the left mid-insula predicted individual differences in JOL accuracy. SPL, superior parietal lobule; Pcu, precuneus;IPL, inferior parietal lobule; IPS, intraparietal sulcus; PG, parahippocampal gyrus; FG, fusiform gyrus; CB, cerebellum.

GM volume of the seed region. Thus, we conducted a two-steplinear regression analysis to examine whether the relationshipbetween JOL accuracy and the functional connectivity of theleft mid-insula was solely due to the correlation between JOL

accuracy and the GM volume of the left mid-insula. JOL accuracywas the outcome variable in the regression analysis. In thefirst step, four control variables (the participants’ age, gender,recognition accuracy, and GM volume of the left mid-insula)

Frontiers in Human Neuroscience | www.frontiersin.org 8 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 9

Hu et al. Judgments of Learning and Insula

FIGURE 4 | The functional connectivity between the six regions observed inthe present study. The parameter for each functional connectivity representsthe correlation between the functional connectivity and the participants’ JOLaccuracy. SPL, superior parietal lobule; Pcu, precuneus; IPL, inferior parietallobule; IPS, intraparietal sulcus; PG, parahippocampal gyrus; FG, fusiformgyrus; CB, cerebellum.

were entered in the model. In the second step, the z-transformedcorrelation values for the functional connectivity between theleft mid-insula and the five clusters were entered in the model.The results revealed that the z-transformed correlation values forthe functional connectivity could explain the additional uniquevariance of JOL accuracy [Fchange (5,25) = 7.892, p < 0.001],which suggested that the functional connectivity of the left mid-insula could predict JOL accuracy independently even whencontrolling for the GM volume of the left mid-insula.

ROI-Wise Functional ConnectivityAnalysis ResultsWe then conducted an exploratory ROI-wise analysis using thefive regions found in the seed-based analysis as ROIs to explorewhether individual differences in JOL accuracy were correlatedwith the functional connectivity between the regions connectedto the left mid-insula. JOL accuracy showed significant positivecorrelations with the z-transformed correlation values of fourROI-to-ROI pairs (see Figure 4): left SPL/Pcu – right frontalpole (r = 0.493, FDR-corrected p = 0.041), left SPL/Pcu – rightIPL/IPS (r = 0.437, FDR-corrected p = 0.041), left SPL/Pcu –left IPL/IPS (r = 0.421, FDR-corrected p = 0.041), and leftIPL/IPS – right IPL/IPS (r = 0.453, FDR-corrected p = 0.041).In addition, none of the ROI-to-ROI functional connectivitywas significantly correlated with recognition accuracy (allp > 0.3).

TABLE 2 | Factor loadings for each of the functional connectivity related to JOLaccuracy (Loadings < 0.30 suppressed).

Functional connectivity Factor

1 2

L. mid-insula – R. frontal pole 0.916

L. mid-insula – L. SPL/Pcu 0.834

L. mid-insula – R. IPL/IPS 0.915

L. mid-insula – L. IPL/IPS 0.952

L. mid-insula – L. PG/FG/CB 0.620

L. SPL/Pcu – R. frontal pole 0.533

L. IPL/IPS – R. IPL/IPS 0.324 0.461

L. SPL/Pcu – L. IPL/IPS 0.871

L. SPL/Pcu – R. IPL/IPS 0.870

L, left; R, right; SPL, superior parietal lobule; Pcu, precuneus; IPL, inferior parietallobule; IPS, intraparietal sulcus; PG, parahippocampal gyrus; FG, fusiform gyrus;CB, cerebellum.

Exploratory Factor Analysis ResultsWe conducted an exploratory factor analysis on the functionalconnectivity of nine ROI-to-ROI pairs related to JOL accuracy(five found in the seed-based analysis and four found in the ROI-wise analysis) to examine whether the functional connectivityfound in the analyses above could form different subsets. Theresults from the exploratory factor analysis revealed that twofactors with an eigenvalue over 1 were extracted, accounting for66.507% of the variance. Factor 1 had higher loadings on thefunctional connectivity of six ROI-to-ROI pairs: left mid-insula- right frontal pole, left mid-insula - left SPL/Pcu, left mid-insula – right IPL/IPS, left mid-insula – left IPL/IPS, left mid-insula – left parahippocampal gyrus/fusiform gyrus/cerebellum,and left SPL/Pcu – right frontal pole. Factor 2 had higher loadingson the functional connectivity of three ROI-to-ROI pairs: leftSPL/Pcu – right IPL/IPS, left SPL/Pcu – left IPL/IPS, and leftIPL/IPS – right IPL/IPS (see Table 2 and Figure 4). In addition,the functional connectivity between the left and right IPL/IPS hadfactor loadings higher than 0.3 on both factors. We also foundthat the correlation between the scores of the two factors wassignificant, r(35)= 0.462, p= 0.005.

To examine whether these two factors could independentlypredict individual differences in JOL accuracy, we conducted athree-step linear regression analysis with JOL accuracy as theoutcome variable. In the first step, the participants’ age, genderand recognition accuracy were entered in the model as controlvariables. These variables could not significantly predict JOLaccuracy, R2

= 0.044, adjusted R2= −0.049, F(3,31) = 0.473,

p = 0.703. In the second and third steps, the scores ofFactors 1 and 2, respectively, were entered in the model. Theresults revealed that the scores of both factors could explainadditional unique variance of JOL accuracy [Factor 1: Fchange(1,30) = 35.966, p < 0.001; Factor 2: Fchange (1,29) = 4.382,p = 0.045], suggesting that both factors could predict JOLaccuracy independently while controlling for age, gender andrecognition accuracy [final regression model: R2

= 0.622,adjusted R2

= 0.557; F(5,29)= 9.553, p < 0.001; age B=−0.012,95% CI [−0.021, −0.003]; gender B = 0.026, 95% CI [−0.012,

Frontiers in Human Neuroscience | www.frontiersin.org 9 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 10

Hu et al. Judgments of Learning and Insula

0.065]; recognition accuracy B=−0.088, 95% CI [−0.283, 0.108];Factor 1 B= 0.046, 95% CI [0.026, 0.066]; Factor 2 B= 0.021, 95%CI [0, 0.042]].

DISCUSSION

The present study investigated the neural correlates of individualdifferences in JOL accuracy. Participants completed a memorytask in which they made JOLs about their recognitionperformance, and we used structural MRI and resting-state fMRIto examine whether individual differences in JOL accuracy werecorrelated with the GM volume and the resting-state functionalconnectivity in the brain. We found that the GM volume ofthe left mid-insula could predict individual differences in theparticipants’ JOL accuracy. The participants’ JOL accuracy wasalso correlated with the resting-state functional connectivitybetween the left mid-insula and various other brain regions,including the left SPL/Pcu, bilateral IPL/IPS, right frontal poleand left parahippocampal gyrus/fusiform gyrus/cerebellum, andthe functional connectivity related to JOL accuracy could bedivided into two subsets. Our findings provide novel evidencefor the involvement of the left mid-insula and other distributedbrain regions and their functional connectivity in the processesrelated to accurate JOLs and suggest that these regions may playan important role in the JOL process.

Our study found that greater JOL accuracy was related tohigher GM volume of the left mid-insula. Our results areconsistent with previous FOK studies indicating that greater FOKaccuracy is correlated with higher GM volume of the insula(Cosentino et al., 2015; Le Berre et al., 2016). These studiessuggest that in FOK processes, the insula may be related toperformance monitoring (Cosentino et al., 2015) or to self-inferential processes for generating accurate future estimations(Le Berre et al., 2016). JOL and FOK are both prospectivemetamemory judgments and involve similar cognitive processes(Dunlosky and Metcalfe, 2009). Thus, the insula may play similarroles in FOK and JOL. In the JOL process, the insula maybe associated with the monitoring of the performance in thememory-encoding phase and the generation of the predictionsfor future memory test. In contrast to previous FOK studies thatused ROI-based approach to analyze the GM volume of the insula(Cosentino et al., 2015; Le Berre et al., 2016), this study usedVBM analysis and found that the GM volume of the left mid-insula was specifically correlated with JOL accuracy. Previousstudies have indicated that the mid-insula plays an importantrole in the evaluation of bodily state (i.e., interoception) (Craig,2011). In the interoceptive process, the mid-insula re-representsthe primary neural representations of the state of the body andintegrates the representations of the body with other neuralinputs to form a combined representation of the individual’sinternal and external environment (Craig, 2009, 2011). Thus, it ispossible that in the JOL process, the left mid-insula may monitorand process the neural representations of different information orcues related to the performance in the memory-encoding phaseand then integrate the cues to generate estimations for futurememory test (Koriat, 1997; Kao et al., 2005). In addition, Chua

et al. (2009) showed that metamemory process was associatedwith less activity in brain regions related to the processingof external visual stimuli, and with greater activity in regionsresponsible for internally directed cognition. Chua et al. (2009)suggest that metamemory is characterized by both a shift towardinternally directed cognition and away from externally directedcognition. One possibility is that the left mid-insula may serve as ahub between external representation and internal representationand may be related to the shift toward internally directedrepresentation. Moreover, Critchley et al. (2004) suggest that inthe interoceptive process, the insula is related to the subjectivefeeling states that may underlie the conscious representation ofour internal bodily processes. Perhaps the left mid-insula is alsoassociated with the conscious representation of our experiencein the memory-encoding phase. Future research should furtherexamine the role of the left mid-insula in the JOL process.

One difference between this study and previous FOK studies isthat previous studies have indicated that FOK accuracy is relatedto the right or bilateral insula (Cosentino et al., 2015; Le Berreet al., 2016), whereas this study revealed a significant relationshipbetween JOL accuracy and the left mid-insula. One possibleexplanation is that the insula in different hemispheres might berelated to prospective metamemory judgments at different stages:JOL is made in the encoding phase, while FOK is made duringthe memory test (Dunlosky and Metcalfe, 2009). In addition, wealso note that results from previous FOK studies are not identical(Cosentino et al., 2015; Le Berre et al., 2016), and it is possible thatthe different tasks used in this study and previous studies maylead to differences in lateralization. For example, Cosentino et al.(2015) asked participants to learn information about differentpeople and make FOK judgments. They found that FOK accuracywas related to the right insular volume. By contrast, Le Berreet al. (2016) required participants to learn unrelated word pairsand found that the FOK accuracy was related to the GM volumeof the bilateral insula. In contrast to previous studies, this studyrequired participants to learn a list of words and perform arecognition test. Future research should use VBM analysis toexamine the relationship between the GM volume in the insulaand both types of prospective metamemory judgments in a singleexperiment.

Contrary to our hypothesis, individual differences in JOLaccuracy were not related to the GM volume of medial BA 11,which is activated in the JOL process according to previous task-related fMRI studies (Kao et al., 2005; Do Lam et al., 2012;Yang et al., 2015). In addition, our results from resting-statefMRI analysis showed that the functional connectivity of medialBA 11 was not correlated with JOL accuracy. Structural MRIand resting-state fMRI analyses aim to reveal the variability inbrain volume or resting-state activity in relevant neural structuresthat reflect individual differences in performance. It does notstrictly follow that the identified region is more active duringthe relevant tasks, because the relationship between task-relatedfMRI findings and the results of structural MRI and resting-statefMRI analyses is complex (Mennes et al., 2010; Kanai and Rees,2011; McCurdy et al., 2013). Future research should combinestructural MRI, resting-state fMRI and task-related fMRI tofurther investigate the neural correlates of the JOL process. In

Frontiers in Human Neuroscience | www.frontiersin.org 10 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 11

Hu et al. Judgments of Learning and Insula

addition, even in previous task-related fMRI studies concerningJOL, the direction of correlation between JOL accuracy andthe activation of medial BA 11 has been controversial (Kaoet al., 2005; Yang et al., 2015). For example, Kao et al. (2005)required participants to learn pictures of scenes and make JOLsfor all pictures. They found that JOL accuracy showed a positivecorrelation with activation in medial BA 11. In contrast, Yanget al. (2015) asked participants to learn a list of words andmake JOLs for only half of the words. Their results revealed thatgreater JOL accuracy was correlated with a smaller subsequentJOL effect (i.e., difference in activation between trials given highand low JOLs) in medial BA 11. One possible explanation forthese conflicting results is that the different tasks used in differentstudies may affect the relationship between JOL accuracy andmedial BA 11. Future studies should examine the effect of thetask type on the relationship between JOL accuracy and medialBA 11.

Furthermore, our resting-state fMRI analysis showed thatJOL accuracy was associated with a resting-state network thatincluded the functional connectivity between the left mid-insulaand different regions. These results are consistent with a previousstudy indicating that the mid-insula is functionally connectedwith various brain regions (Cauda et al., 2011). For example,Cauda et al. (2011) indicated that the anterior and posteriorinsula are involved in two different brain networks: the anteriorinsula is functionally connected to the frontal cortex, anteriorcingulate cortex, parietal cortex and superior temporal gyrus,whereas the posterior insula is functionally connected to thesensorimotor cortex, the supplementary motor cortex and severalregions in the temporal cortex and limbic system. In addition,Cauda et al. (2011) suggested that the mid-insula is a transitionalarea between the anterior and posterior insula and that the twonetworks overlap in the mid-insula. Thus, the mid-insula may beconnected to regions in both networks. Furthermore, our resultsrevealed that the functional connectivity found in this study couldbe divided into two subsets or factors (Factors 1 and 2; see Table 2and Figure 4). The functional connectivity within one factor wasmore closely correlated with each other than with the functionalconnectivity in the other factor, and the functional connectivitywithin a single factor may share similar functions (Jacobs et al.,2014). In addition, the two factors could independently predictindividual differences in JOL accuracy, and a higher score of onefactor was significantly correlated with a higher score of the other.

According to our factor analysis, Factor 1 contains thefunctional connectivity between the left mid-insula and variousregions (including the left parahippocampal gyrus/fusiformgyrus/cerebellum, left SPL/Pcu, bilateral IPL/IPS and right frontalpole) and between the right frontal pole and the left SPL/Pcu.The parahippocampal gyrus is an important part of the medialtemporal lobe and plays a key role in the memory encodingof various stimuli (Kirchhoff et al., 2000; Strange et al., 2002).The fusiform gyrus is related to the encoding of phonological orlexical features into episodic memory (Kirchhoff et al., 2000). Inaddition, the functional connectivity between the left mid-insulaand left parahippocampal gyrus/fusiform gyrus is likely related tothe posterior insula network, which may play an important role insensorimotor integration (Cauda et al., 2011). It is possible that in

the JOL process, the functional connectivity between the left mid-insula and the parahippocampal gyrus/fusiform gyrus may alsobe related to the integration of information about the encodingphase. In the JOL process, the left mid-insula may integrate theinformation about the phonological or lexical attributes of words(from the fusiform gyrus) and other information in the encodingprocess (from the parahippocampal gyrus), and higher efficacyof this information integration process may predict higher JOLaccuracy. In addition, the parietal regions found in the presentstudy (including the SPL, Pcu, IPL, and IPS) are closely associatedwith episodic memory (Wagner et al., 2005; Baird et al., 2013),and previous studies have revealed that regions in the dorsalparietal cortex, such as the SPL, Pcu, and IPS, are related totop-down attention in memory encoding (Cabeza et al., 2008;Uncapher et al., 2011). Moreover, the functional connectivitybetween the left mid-insula and left SPL/Pcu, bilateral IPL/IPSand right frontal pole may be related to the anterior insulanetwork, which is associated with attention control (Cauda et al.,2011). The relationship between JOL accuracy and functionalconnectivity between these regions suggests that the controlof our attention in the JOL process may support accurateJOLs. In addition, given the possible role of the mid-insulain the information integration process (Craig, 2011), anotherexplanation for the role of the functional connectivity betweenthe left mid-insula and these regions in the JOL process isthat when JOLs are made, the left mid-insula may receive andintegrate the information about attention control from theseregions.

Moreover, the frontal pole has also been shown to be involvedin the JOL process. For example, Ryals et al. (2016) found thattranscranial magnetic stimulation (TMS) on the frontal polecould significantly increase JOL accuracy. The frontal pole isthe apex of the rostro-caudal hierarchically organized prefrontalcortex (Badre, 2008; Badre and D’Esposito, 2009), and Ryalset al. (2016) suggest that the frontal pole may interact with othermemory-processing regions in the JOL process and integrateinformation from these posterior regions. Thus, the functionalconnectivity between the right frontal pole and left SPL/Pcu maybe associated with the integration of information in the JOLprocess. Taken together, the functional connectivity in Factor 1may reflect the process of information integration, which is likelyto support accurate JOLs.

In our factor analysis, Factor 2 contains the functionalconnectivity between the left SPL/Pcu and bilateral IPL/IPS.These regions mainly belong to the dorsal attention network(DAN) and the frontoparietal network (FPN) (Yeo et al.,2011). The DAN and FPN support the top-down control ofattention and goal-directed processing (Dosenbach et al., 2007;Cabeza et al., 2008). Previous research has shown that selectiveattention toward the to-be-remembered stimuli can enhance theprocessing of the stimuli (Uncapher et al., 2011), which may havepromoted JOL accuracy in the present study. Our results indicatethat JOL accuracy may be related to the functional connectivitywithin the DAN and FPN, suggesting a possible role of selectiveattention in making accurate JOLs. Interestingly, the functionalconnectivity between the bilateral IPL/IPS in Factor 2 also hada factor loading higher than 0.3 on Factor 1, suggesting that

Frontiers in Human Neuroscience | www.frontiersin.org 11 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 12

Hu et al. Judgments of Learning and Insula

this functional connectivity may be related to the integration ofinformation in the JOL process. Future studies should furtherinvestigate the role of the bilateral IPL/IPS in the process ofinformation integration when people make JOLs.

Our behavioral results indicated that individual differencesin JOL accuracy did not correlate with recognition accuracy.In addition, we found that the GM volume and functionalconnectivity of the left mid-insula, which showed significantcorrelation with JOL accuracy, was not related to recognitionaccuracy, suggesting that the cognitive processes underlying JOLaccuracy and memory performance might be distinct. Recentstudies have obtained similar results with brain stimulationtechniques (Chua and Ahmed, 2016; Ryals et al., 2016). Forexample, Ryals et al. (2016) found that TMS on the frontal polecould increase JOL accuracy but showed no effect on recognitionperformance. These results are also consistent with previousfunctional MRI studies showing that different brain regionsare activated during memory process and JOL, indicating adissociation between the two processes (Kao et al., 2005; Do Lamet al., 2012). According to the cue-utilization theory, participantsdo not base their JOLs on direct assessments of memory strength.Instead, their JOLs are based on various cues obtained in theencoding phase, and JOL accuracy depends on whether thesecues can accurately predict memory performance (Koriat, 1997;Dunlosky and Metcalfe, 2009). Thus, JOL accuracy reflects thecorrespondence between JOL and memory performance (ratherthan memory performance itself), and the brain regions that areessential for JOL accuracy may not be involved in the memoryprocess. Future studies should further disentangle the neuralcorrelates of memory performance and JOL accuracy.

Our study has several limitations. First, the sample size(35 participants) in the present study was not large enough.Second, similar to all VBM and resting-state functionalconnectivity studies, our results are correlational in nature anddo not imply any direct causality between individual differencesin JOL accuracy and the GM volume of the left mid-insulaor resting-state functional connectivity. Although individualdifferences in brain structure and resting-state activity are closelylinked to differences in brain function (Kanai and Rees, 2011;Lee et al., 2013), more evidence is needed to further examinethe results of this study and our explanation concerning theneural mechanisms related to individual differences in JOLaccuracy. Third, our conclusions are not exhaustive, as theydo not exclude the possibility that other regions may also berelated to JOL accuracy without any difference in the GM volume

or resting-state activity. Finally, we did not investigate whetherindividual differences in JOL accuracy are related to othercognitive functions. For example, a previous study suggestedthat executive function may be related to metamemory processes(Destan and Roebers, 2015). Future studies should use largersample sizes and combine event-related fMRI, structural MRIand resting-state fMRI techniques to further examine the neuralmechanisms concerning individual differences in JOL accuracy.In addition, future studies should also investigate the similaritiesand differences between the neural mechanisms of individualdifferences in JOL accuracy and other cognitive functions, suchas executive function.

ETHICS STATEMENT

This study was carried out in accordance with therecommendations of the Institutional Review Board of theState Key Laboratory of Cognitive Neuroscience and Learning,Beijing Normal University with written informed consent fromall subjects. All subjects gave written informed consent inaccordance with the Declaration of Helsinki.

AUTHOR CONTRIBUTIONS

XH and LL designed the experiment. XH, WC, JZ, NS, WW, andLL performed the experiment. XH, ZL, and LL analyzed the data.XH, ZL, CL, and LL wrote the manuscript.

FUNDING

This work was supported by the National Natural ScienceFoundation of China (31671130), the Humanity and SocialScience Foundation (20116030) and the Training ProgramFoundation for the Excellent Youth Scholars of China Universityof Political Science and Law.

ACKNOWLEDGMENT

We thank all participants who volunteered to take part inthis study and the reviewers for their insightful comments andsuggestions.

REFERENCESAshburner, J. (2007). A fast diffeomorphic image registration algorithm.

Neuroimage 38, 95–113. doi: 10.1016/j.neuroimage.2007.07.007Badre, D. (2008). Cognitive control, hierarchy, and the rostro-caudal organization

of the frontal lobes. Trends Cogn. Sci. 12, 193–200. doi: 10.1016/j.tics.2008.02.004

Badre, D., and D’Esposito, M. (2009). Is the rostro-caudal axis of the frontal lobehierarchical? Nat. Rev. Neurosci. 10, 659–669. doi: 10.1038/nrn2667

Baird, B., Smallwood, J., Gorgolewski, K. J., and Margulies, D. S. (2013). Medialand lateral networks in anterior prefrontal cortex support metacognitive ability

for memory and perception. J. Neurosci. 33, 16657–16665. doi: 10.1523/JNEUROSCI.0786-13.2013

Beaman, P., Hanczakowski, M., and Jones, D. M. (2014). The effects ofdistraction on metacognition and metacognition on distraction: evidencefrom recognition memory. Front. Psychol. 5:439. doi: 10.3389/fpsyg.2014.00439

Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate:a practical and powerful approach to multiple testing. J. R. Stat. Soc. B 57,289–300. doi: 10.2307/2346101

Bettman, J. R., Luce, M. F., and Payne, J. W. (1998). Constructive consumer choiceprocesses. J. Consum. Res. 25, 187–217. doi: 10.1086/209535

Frontiers in Human Neuroscience | www.frontiersin.org 12 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 13

Hu et al. Judgments of Learning and Insula

Bhaskar, S., Saeidi, K., Borhani, P., and Amiri, H. (2013). Recent progressin migraine pathophysiology: role of cortical spreading depression andmagnetic resonance imaging. Eur. J. Neurosci. 38, 3540–3551. doi: 10.1111/ejn.12368

Bjork, R. A., Dunlosky, J., and Kornell, N. (2013). Self-regulated learning: beliefs,techniques, and illusions. Annu. Rev. Psychol. 64, 417–444. doi: 10.1146/annurev-psych-113011-143823

Cabeza, R., Ciaramelli, E., Olson, I. R., and Moscovitch, M. (2008). The parietalcortex and episodic memory: an attentional account. Nat. Rev. Neurosci. 9,613–625. doi: 10.1038/nrn2459

Cai, Q., and Brysbaert, M. (2010). SUBTLEX-CH: Chinese word and characterfrequencies based on film subtitles. PLoS ONE 5:e10729. doi: 10.1371/journal.pone.0010729

Cauda, F., D’Agata, F., Sacco, K., Duca, S., Geminiani, G., and Vercelli, A. (2011).Functional connectivity of the insula in the resting brain. Neuroimage 55, 8–23.doi: 10.1016/j.neuroimage.2010.11.049

Chua, E. F., and Ahmed, R. (2016). Electrical stimulation of the dorsolateralprefrontal cortex improves memory monitoring. Neuropsychologia 85, 74–79.doi: 10.1016/j.neuropsychologia.2016.03.008

Chua, E. F., Pergolizzi, D., and Weintraub, R. R. (2014). “The cognitiveneuroscience of metamemory monitoring: understanding metamemoryprocesses, subjective levels expressed, and metacognitive accuracy,” in TheCognitive Neuroscience of Metacognition, eds S. Fleming and C. Frith (Berlin:Springer), 267–291. doi: 10.1007/978-3-642-45190-4_12

Chua, E. F., Schacter, D. L., and Sperling, R. A. (2009). Neural correlatesof metamemory: a comparison of feeling-of-knowing and retrospectiveconfidence judgments. J. Cogn. Neurosci. 21, 1751–1765. doi: 10.1162/jocn.2009.21123

Cosentino, S., Brickman, A. M., Griffith, E., Habeck, C., Cines, S., Farrell, M., et al.(2015). The right insula contributes to memory awareness in cognitively diverseolder adults. Neuropsychologia 75, 163–169. doi: 10.1016/j.neuropsychologia.2015.05.032

Craig, A. D. (2009). Emotional moments across time: a possible neural basis fortime perception in the anterior insula. Philos. Trans. R. Soc. Lond. B Biol. Sci.364, 1933–1942. doi: 10.1098/rstb.2009.0008

Craig, A. D. (2011). Significance of the insula for the evolution of human awarenessof feelings from the body.Ann. N. Y. Acad. Sci. 1225, 72–82. doi: 10.1111/j.1749-6632.2011.05990.x

Critchley, H. D., Wiens, S., Rotshtein, P., Öhman, A., and Dolan, R. J. (2004).Neural systems supporting interoceptive awareness. Nat. Neurosci. 7, 189–195.doi: 10.1038/nn1176

Destan, N., and Roebers, C. M. (2015). What are the metacognitive costs of youngchildren’s overconfidence? Metacogn. Learn. 10, 347–374. doi: 10.1007/s11409-014-9133-z

Do Lam, A. T. A., Axmacher, N., Fell, J., Staresina, B. P., Gauggel, S.,Wagner, T., et al. (2012). Monitoring the mind: the neurocognitivecorrelates of metamemory. PLoS ONE 7:e30009. doi: 10.1371/journal.pone.0030009

Dosenbach, N. U. F., Fair, D. A., Miezin, F. M., Cohen, A. L., Wenger, K. K.,Dosenbach, R. A. T., et al. (2007). Distinct brain networks for adaptive andstable task control in humans. Proc. Natl. Acad. Sci. U.S.A. 104, 11073–11078.doi: 10.1073/pnas.0704320104

Dunlosky, J., and Metcalfe, J. (2009). Metacognition. Thousand Oaks, CA: Sage.Evans, T. M., Kochalka, J., Ngoon, T. J., Wu, S. S., Qin, S., Battista, C., et al.

(2015). Brain structural integrity and intrinsic functional connectivity forecast6 year longitudinal growth in children’s numerical abilities. J. Neurosci. 35,11743–11750. doi: 10.1523/JNEUROSCI.0216-15.2015

Fabrigar, L. R., Wegener, D. T., MacCallum, R. C., and Strahan, E. J. (1999).Evaluating the use of exploratory factor analysis in psychological research.Psychol. Methods 4, 272–299. doi: 10.1037/1082-989X.4.3.272

Fleming, S. M., and Lau, H. C. (2014). How to measure metacognition. Front. Hum.Neurosci. 8:443. doi: 10.3389/fnhum.2014.00443

Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., and Rees, G. (2010). Relatingintrospective accuracy to individual differences in brain structure. Science 329,1541–1543. doi: 10.1126/science.1191883

Fox, M. D., and Raichle, M. E. (2007). Spontaneous fluctuations in brain activityobserved with functional magnetic resonance imaging. Nat. Rev. Neurosci. 8,700–711. doi: 10.1038/nrn2201

Friston, K. J., Williams, S., Howard, R., Frackowiak, R. S. J., and Turner, R. (1996).Movement-related effects in fMRI time-series. Magn. Reson. Med. 35, 346–355.doi: 10.1002/mrm.1910350312

Galvin, S. J., Podd, J. V., Drga, V., and Whitmore, J. (2003). Type 2 tasks inthe theory of signal detectability: discrimination between correct and incorrectdecisions. Psychon. Bull. Rev. 10, 843–876. doi: 10.3758/BF03196546

Hayasaka, S., Phan, K. L., Liberzon, I., Worsley, K. J., and Nichols, T. E. (2004).Nonstationary cluster-size inference with random field and permutationmethods. Neuroimage 22, 676–687. doi: 10.1016/j.neuroimage.2004.01.041

Hogeveen, J., Hauner, K. K., Chau, A., Krueger, F., and Grafman, J. (2016).Impaired valuation leads to increased apathy following ventromedial prefrontalcortex damage. Cereb. Cortex 27, 1401–1408. doi: 10.1093/cercor/bhv317

Jacobs, R. H., Jenkins, L. M., Gabriel, L. B., Barba, A., Ryan, K. A., Weisenbach, S. L.,et al. (2014). Increased coupling of intrinsic networks in remitted depressedyouth predicts rumination and cognitive control. PLoS ONE 9:e104366.doi: 10.1371/journal.pone.0104366

Kanai, R., and Rees, G. (2011). The structural basis of inter-individual differencesin human behaviour and cognition. Nat. Rev. Neurosci. 12, 231–242. doi: 10.1038/nrn3000

Kao, Y. C., Davis, E. S., and Gabrieli, J. D. E. (2005). Neural correlates of actual andpredicted memory formation.Nat. Neurosci. 8, 1776–1783. doi: 10.1038/nn1595

Kirchhoff, B. A., Wagner, A. D., Maril, A., and Stern, C. E. (2000). Prefrontal–temporal circuitry for episodic encoding and subsequent memory. J. Neurosci.20, 6173–6180.

Koriat, A. (1997). Monitoring one’s own knowledge during study: a cue-utilizationapproach to judgments of learning. J. Exp. Psychol. Gen. 126, 349–370.doi: 10.1037/0096-3445.126.4.349

Kornbrot, D. E. (2006). Signal detection theory, the approach of choice: model-based and distribution-free measures and evaluation. Percept. Psychophys. 68,393–414. doi: 10.3758/BF03193685

Le Berre, A.-P., Müller-Oehring, E. M., Kwon, D., Serventi, M. R., Pfefferbaum, A.,and Sullivan, E. V. (2016). Differential compromise of prospective andretrospective metamemory monitoring and their dissociable structural braincorrelates. Cortex 81, 192–202. doi: 10.1016/j.cortex.2016.05.002

Lee, M. H., Smyser, C. D., and Shimony, J. S. (2013). Resting-state fMRI: areview of methods and clinical applications. Am. J. Neuroradiol. 34, 1866–1872.doi: 10.3174/ajnr.A3263

Li, N., Ma, N., Liu, Y., He, X. S., Sun, D. L., Fu, X. M., et al. (2013). Resting-state functional connectivity predicts impulsivity in economic decision-making.J. Neurosci. 33, 4886–4895. doi: 10.1523/JNEUROSCI.1342-12.2013

Lim, S.-L., O’Doherty, J. P., and Rangel, A. (2013). Stimulus value signals inventromedial PFC reflect the integration of attribute value signals computedin fusiform gyrus and posterior superior temporal gyrus. J. Neurosci. 33,8729–8741. doi: 10.1523/JNEUROSCI.4809-12.2013

Loh, K. K., and Kanai, R. (2014). Higher media multi-tasking activity is associatedwith smaller gray-matter density in the anterior cingulate cortex. PLoS ONE9:e106698. doi: 10.1371/journal.pone.0106698

Luders, E., Gaser, C., Jancke, L., and Schlaug, G. (2004). A voxel-based approach togray matter asymmetries. Neuroimage 22, 656–664. doi: 10.1016/j.neuroimage.2004.01.032

Masson, M. E., and Rotello, C. M. (2009). Sources of bias in the Goodman-Kruskal gamma coefficient measure of association: implications for studiesof metacognitive processes. J. Exp. Psychol. Learn. Mem. Cogn. 35, 509–527.doi: 10.1037/a0014876

McCurdy, L. Y., Maniscalco, B., Metcalfe, J., Liu, K. Y., de Lange, F. P., andLau, H. (2013). Anatomical coupling between distinct metacognitive systemsfor memory and visual perception. J. Neurosci. 33, 1897–1906. doi: 10.1523/JNEUROSCI.1890-12.2013

Mennes, M., Kelly, C., Zuo, X. N., Di Martino, A., Biswal, B. B., Castellanos, F. X.,et al. (2010). Inter-individual differences in resting-state functional connectivitypredict task-induced BOLD activity. Neuroimage 50, 1690–1701. doi: 10.1016/j.neuroimage.2010.01.002

Nelson, T. O. (1984). A comparison of current measures of the accuracy of feeling-of-knowing predictions. Psychol. Bull. 95, 109–133. doi: 10.1037/0033-2909.95.1.109

Nelson, T. O., and Narens, L. (1990). “Metamemory: a theoretical framework andnew findings,” in Psychology of Learning and Motivation, ed. H. B. Gordon

Frontiers in Human Neuroscience | www.frontiersin.org 13 August 2017 | Volume 11 | Article 399

fnhum-11-00399 July 28, 2017 Time: 16:14 # 14

Hu et al. Judgments of Learning and Insula

(New York, NY: Academic Press), 125–173. doi: 10.1016/S0079-7421(08)60053-5

Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., and Petersen, S. E.(2012). Spurious but systematic correlations in functional connectivity MRInetworks arise from subject motion. Neuroimage 59, 2142–2154. doi: 10.1016/j.neuroimage.2011.10.018

Ryals, A. J., Rogers, L. M., Gross, E. Z., Polnaszek, K. L., and Voss, J. L.(2016). Associative recognition memory awareness improved by theta-burststimulation of frontopolar cortex. Cereb. Cortex 26, 1200–1210. doi: 10.1093/cercor/bhu311

Saad, Z. S., Gotts, S. J., Murphy, K., Chen, G., Jo, H. J., Martin, A., et al.(2012). Trouble at rest: how correlation patterns and group differences becomedistorted after global signal regression. Brain Connect. 2, 25–32. doi: 10.1089/brain.2012.0080

Schneider, W., Eschman, A., and Zuccolotto, A. (2002). E-Prime: User’s Guide.Pittsburgh, PA: Psychology Software Incorporated.

Schnyer, D. M., Nicholls, L., and Verfaellie, M. (2005). The role of VMPCin metamemorial judgments of content retrievability. J. Cogn. Neurosci. 17,832–846. doi: 10.1162/0898929053747694

Schwartz, B. L., and Bacon, E. (2008). “Metacognitive neuroscience,” in Handbookof Memory and Metamemory: Essays in Honor of Thomas O. Nelson, edsJ. Dunlosky and R. A. Bjork (New York, NY: Psychology Press), 355–371.doi: 10.4324/9780203805503.ch18

Schwarz, N. (2004). Metacognitive experiences in consumer judgment anddecision making. J. Consum. Psychol. 14, 332–348. doi: 10.1207/s15327663jcp1404_2

Song, M., Zhou, Y., Li, J., Liu, Y., Tian, L., Yu, C., et al. (2008). Brain spontaneousfunctional connectivity and intelligence. Neuroimage 41, 1168–1176. doi: 10.1016/j.neuroimage.2008.02.036

Strange, B. A., Otten, L. J., Josephs, O., Rugg, M. D., and Dolan, R. J. (2002).Dissociable human perirhinal, hippocampal, and parahippocampal roles duringverbal encoding. J. Neurosci. 22, 523–528.

Uncapher, M. R., Hutchinson, J. B., and Wagner, A. D. (2011). Dissociable effectsof top-down and bottom-up attention during episodic encoding. J. Neurosci. 31,12613–12628. doi: 10.1523/JNEUROSCI.0152-11.2011

Valk, S. L., Bernhardt, B. C., Bockler, A., Kanske, P., and Singer, T. (2016).Substrates of metacognition on perception and metacognition on higher-ordercognition relate to different subsystems of the mentalizing network. Hum. BrainMapp. 37, 3388–3399. doi: 10.1002/hbm.23247

Vilkki, J., Servo, A., and Surma-aho, O. (1998). Word list learning and predictionof recall after frontal lobe lesions. Neuropsychology 12, 268–277. doi: 10.1037/0894-4105.12.2.268

Vilkki, J., Surma-aho, O., and Servo, A. (1999). Inaccurate prediction of retrieval ina face matrix learning task after right frontal lobe lesions. Neuropsychology 13,298–305. doi: 10.1037/0894-4105.13.2.298

Wagner, A. D., Shannon, B. J., Kahn, I., and Buckner, R. L. (2005). Parietal lobecontributions to episodic memory retrieval. Trends Cogn. Sci. 9, 445–453.doi: 10.1016/j.tics.2005.07.001

Xiao, Y., Friederici, A. D., Margulies, D. S., and Brauer, J. (2016).Longitudinal changes in resting-state fMRI from age 5 to age 6 yearscovary with language development. Neuroimage 128, 116–124. doi:10.1016/j.neuroimage.2015.12.008

Yan, C., and Zang, Y. (2010). DPARSF: a MATLAB toolbox for “pipeline” dataanalysis of resting-state fMRI. Front. Syst. Neurosci. 4:13. doi: 10.3389/fnsys.2010.00013

Yan, C.-G., Cheung, B., Kelly, C., Colcombe, S., Craddock, R. C., Di Martino, A.,et al. (2013). A comprehensive assessment of regional variation in the impact ofhead micromovements on functional connectomics. Neuroimage 76, 183–201.doi: 10.1016/j.neuroimage.2013.03.004

Yang, H., Cai, Y., Liu, Q., Zhao, X., Wang, Q., Chen, C., et al. (2015). Differentialneural correlates underlie judgment of learning and subsequent memoryperformance. Front. Psychol. 6:1699. doi: 10.3389/fpsyg.2015.01699

Yeo, B. T. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D.,Hollinshead, M., et al. (2011). The organization of the human cerebral cortexestimated by intrinsic functional connectivity. J. Neurophysiol. 106, 1125–1165.doi: 10.1152/jn.00338.2011

Conflict of Interest Statement: The authors declare that the research wasconducted in the absence of any commercial or financial relationships that couldbe construed as a potential conflict of interest.

Copyright © 2017 Hu, Liu, Chen, Zheng, Su, Wang, Lin and Luo. This is an open-access article distributed under the terms of the Creative Commons AttributionLicense (CC BY). The use, distribution or reproduction in other forums is permitted,provided the original author(s) or licensor are credited and that the originalpublication in this journal is cited, in accordance with accepted academic practice.No use, distribution or reproduction is permitted which does not comply with theseterms.

Frontiers in Human Neuroscience | www.frontiersin.org 14 August 2017 | Volume 11 | Article 399